The QML project at C-DAC Noida focuses on developing Quantum Machine Learning models for weather forecasting using real-world and real-time datasets such as NCMRWF, NASA-POWER, Meteostat, open-meteo, weather api, ERA5, and IMD. The project covers the complete end-to-end workflow including data preprocessing, feature engineering, model training, validation, testing and forecasting with user interface development.

A key highlight of the project is the development of a real-time QML-based Weather Meteogram that provides interactive 10-day forecasts for parameters like Temperature, Humidity and Pressure for major Indian cities. The system compares classical models (LSTM, GRU, ANN) with hybrid Quantum-Classical Machine Learning models (QLSTM, QGRU, HQNN), demonstrating significant parameter reduction while maintaining competitive or superior accuracy.

The project has also conducted detailed noise analysis on quantum circuits and is actively working towards deploying these models on real Rigetti Quantum Processing Units (QPUs) via AWS Braket, moving from simulation to practical quantum hardware execution.

• Real-time QML Weather Meteogram: Daily updated 10-day forecasts for Temperature, Humidity, Pressure, etc., for major Indian cities.
• Hybrid Model Benchmarking: Systematic comparison of Quantum models (QLSTM, QGRU, HQNN) against classical baselines for accuracy, parameter efficiency, and robustness.
• Noise-Resilient Quantum Modeling: Analysis and mitigation of quantum hardware noise (readout, dephasing, amplitude damping) for reliable forecasting.
• Transfer Learning & Multi-City Generalization: Exploring model adaptability across different geographical locations and datasets.
• Edge & QPU Deployment: Future roadmap includes deploying optimized QML models on hybrid CPU-GPU platforms and eventually on real quantum hardware (Rigetti QPU).

• Significant Parameter Reduction: Quantum models achieve at least ~50% parameter reduction compared to classical counterparts while maintaining high accuracy, proving usefulness for edge-inferencing required for real-time forecasting.
• Comprehensive Noise Analysis: Detailed study of quantum errors (readout, dephasing, amplitude damping) and their impact on forecasting reliability.
• End-to-End Pipeline: Complete workflow from raw meteorological data to real-time forecasting and visualization (Real-time forecasting and live Meteogram).
• Hybrid Quantum-Classical Architecture: Optimal balance of quantum expressivity and classical trainability for near-term hardware.
• Real-time Capability: Designed for daily updated forecasts, suitable for operational meteorological applications.

• Quantum Models: QLSTM, QGRU, HQNN with Variational Quantum Circuits (Angle Embedding + Strongly Entangling Layers).
• Classical Baselines: LSTM, GRU, ANN for comparative analysis.
• Primary Dataset: NCMRWF Reanalysis / NASA-POWER for training-validation-testing and one-time forecasting; combination of NASA-POWER/meteosource/open-meteo/meteostat for automated fine-tune and real-time forecast generation.
• Key Features: Lagged variables, rolling statistics, cyclical time encoding (sin/cos DOY).
• Hardware Target: Hybrid CPU-GPU platforms; future deployment on AWS Braket Rigetti QPU (Cepheus-1-108Q).
• Performance Metrics: RMSE, MAE, R², with special focus on parameter efficiency and noise resilience.

Software Stack

• Python ecosystem: PyTorch, PennyLane, Qiskit, TensorFlow/Keras
• AWS Braket SDK (for access to Rigetti Cepheus-108 Qubits QPU)
• NVIDIA CUDA 12.2 (for GPU-accelerated training and simulation)

Hardware Stack

• CPU-GPU: NVIDIA RTX A6000 (48 GB VRAM) + Dual Intel Xeon Gold 6430 (128 threads), Main 256 GB RAM
• Storage: 3.7 TB HDD (archival) + 1.8 TB NVMe SSD

Live Weather API Data Access

• NCMRWF / IMD Data Portals: Primary source for operational Indian meteorological data
• Public APIs: NASA POWER, Meteostat, Open-Meteo, Meteosource and weatherapi APIs used for backup, validation, and multi-city generalization

Live Weather API Ingestion

• Primary Portals: NCMRWF and IMD Data Portals for operational Indian meteorological data streams.
• Backup & Verification Arrays: Public API access points including NASA POWER, Meteostat, Open-Meteo, Meteosource, and WeatherAPI.

Abhishek Tiwari, Scientist F

Embedded Systems Group,

C-DAC, Noida

Email: abhishek[at]cdac[dot]in